Introduction

One of the most important implications of the classic papers of Solow (1956) and Swan (1956) is that the lower the starting level of real per capita GDP, relative to the long run of steady state position, the faster is the growth rate. The Solow–Swan model assumes a constant-returns-to-scale production function with two inputs, capital and labor, and substitution between inputs, a constant savings rate, constant rate of growth of population, and neutral technical change, all exogenously given. Convergence of economies starting out at different levels of per capita income to the same steady state rate of growth refflects the diminishing returns to capital implied by the production function assumed: economies starting out with lower levels of real per capita GDP relative to the long-run or steady state position have less capital per worker and therefore higher rates of return to capital. I will refer to this as the standard Barro-Baumol (BB) sense of the meaning of convergence. There is a good deal of current discussion regarding the appropriate deffinition of “convergence.” My purpose here is not to question this notion of convergence but rather to show that estimates of the coefficient of the lagged dependent variable in a dynamic panel model which has been used to study this phenomenon are extremely sensitive to the method of estimation employed.